Nighttime streetscapes are critical to perceived safety and urban vitality, yet their perceptual quality remains difficult to quantify for planning and renewal. This study quantifies how image-derived visual metrics relate to public perception of nighttime streetscapes in Nanjing’s Old City, China, and develops a street-level perception mapping framework. A total of 3,365 valid nighttime panoramic images were collected along the main road network. Fourteen objective visual metrics, covering brightness, color, and scene attributes, were computed for each image and mapped in GIS. Public perception of safety and aesthetics was assessed through a laboratory-based survey of selected images. Correlation analysis and stepwise regression were used to identify systematic relationships between visual metrics and perception. A Vision Transformer-based regression model was further trained to predict public perception of nighttime streetscapes. Together, these analyses demonstrate how interpretable visual metrics and data-driven modeling can be combined to quantify perceptual variation and support evidence-based diagnosis of nighttime street environments in historic urban areas.